Your task in this exercise is to create an R Markdown document that reports an analysis based on the (synthetic) data from the GESIS Panel Special Survey on the Coronavirus SARS-CoV-2 Outbreak in Germany. The output should be an HTML document. For the R Markdown output (and potentially also other types of output you may produce), we suggest that you create a separate folder in your project directory named output.

In the R Markdown documents we will create in this session and the one on Advanced R Markdown & LaTeX, we will use a few packages in addition to the tidyverse. For creating R Markdown documents, you need the rmarkdown package. You can check whether you have a certain package installed via the Packages tab in the RStudio GUI. Alternatively, you can also run the following command (in the R console): "rmarkdown" %in% rownames(installed.packages()). We will also need the following packages: knitr(should have been installed with RStudio), corrr for computing correlations, and stargazer for creating tables. As a reminder, you can install multiple packages via the command install.packages(c("corrr", "stargazer")).

Please note that the analyses we propose here are only toy examples. If you are or feel familiar enough with R and R Markdown and are interested in exploring other questions with the data, feel free to do something else in the R Markdown documents (e.g., use other variables and/or analysis methods). Also feel free to adjust/change anything else in the solution code (e.g., the document title, formatting, structure/headings, etc.).

We have created a sample solution for the R Markdown report that you should create in this exercise. You can find the .Rmd file as well as the HTML output in the solutions folder within the workshop materials: Report1_Obeying_Curfew.Rmd and Report1_Obeying_Curfew.html.

As a general note: The code shown in the solution fields for this set of exercises needs to be included in code chunks in the R Markdown document for which you may also need (or want) to specify (extra) chunk options.

That being said, let’s get started with creating the reproducible R Markdown report. In this report we want to look at factors predicting willingness to comply with curfew measures during early phases of the COVID-19 pandemic in Germany.

1

Create a new R Markdown file in the src folder in your project directory. The output type should be HTML. Also specify a meaningful title and a subtitle, add an author name, and a date in the YAML header.

You can create a new R Markdown document via the File tab in the RStudio menu: File -> New File -> R Markdown. Make sure to select HTML as the output format. You can also specify a title and add your name into the Author field in the GUI menu. You can manually specify a date in the YAML header or insert (inline) R code for automatically pasting the current date.

solution 

---
title: "Factors predicting compliance with curfew measures during early phases of the COVID-19 pandemic"
subtitle: "An analysis based on the GESIS Panel Special Survey on the Coronavirus SARS-CoV-2 Outbreak in Germany"
author: "R user"
date: "`r Sys.Date()`"
output: 
  html_document
---

Optional

If you want to, you can also make your HTML output easier to use and also prettier by specifying some additional options in the YAML header. To increase the usability of the resulting HTML document, you can add a table of contents (TOC) with 2 levels, make the TOC float (i.e., move when you scroll through the document), number the sections (as specified by the headers), hide the code by default, and include a button for downloading the underlying .Rmd document.

For styling the document, our suggestion is to use the Flatly Bootswatch theme and the Tango Pandoc syntax highlighting style (but feel free to use a different theme and/or code highlighting style).

You need to specify the additional options for document functions, structure, and styling as suboptions under the html_document specification. Have a look at the lecture slides for an example of how to do this.

solution 

---
title: "Factors predicting compliance with curfew measures during early phases of the COVID-19 pandemic"
subtitle: "An analysis based on the GESIS Panel Special Survey on the Coronavirus SARS-CoV-2 Outbreak in Germany"
author: "R user"
date: "`r Sys.Date()`"
output: 
  html_document:
    toc: true
    toc_depth: 2
    number_sections: true
    toc_float: true
    code_folding: hide
    code_download: true
    theme: flatly
    highlight: tango
---

2

The next setup step is setting some knitr code chunk options. Make sure that the code is displayed in the output document, but that messages are suppressed.

You probably do not want to include the chunk options as code output in the resulting HTML document. You can achieve this by setting the chunk option include to FALSE for the setup chunk.

solution

{r setup, include=FALSE} knitr::opts_chunk$set(echo = TRUE, message = FALSE)

NB: In your R Markdown document, this code block needs to start and end with three backticks. Remember that in RStudio you can use the keyboard shortcut Ctrl + Alt + I (Windows & Linux)/Cmd + Option + I (Mac) for inserting code chunks.

3

As a final setup step for the document and the results to be reported therein, you should load the required packages and go through the data wrangling steps included in the R script we have created in the previous session.

You can either copy the code from the wrangling script we have created before (which should be called data_wrangling_gpc.R and stored in the src folder within your project repository) or you can call the script from within the R Markdown document. The command for executing an R script from another R script or an R Markdown document is source() which takes as its first argument a string containing the path to the file.

solution 

source("data_wrangling_gpc.R")

library(corrr)
library(stargazer)

4

Now that we have set up the document, let’s start generating some content for it. We propose the following structure (but, again, feel free to change/adapt this): Research question, Methods with the two subsections Sample and Measure, Results with the three subsections Descriptive statistics, Correlations, and Regression analysis, and a final Discussion section.

You can already add some text to the Research question section.

You can use the Markdown conventions for specifying different header levels using the # symbol.

solution

The number of # symbols is equivalent to the header level.

5

In the Sample (sub)section, indicate how many respondents the data set includes and how many of them are male and female. For our report, we will only use the responses from people who have not indicated that they work in critical professions. You should have created this filtered data set as part of the data wrangling script. Also report in this section that we will use this reduced data set as well as how many respondents it includes and how many of those are male and female.

You can use inline code for dynamically producing the required numbers within the text. Remember that inline code needs to be enclosed in single backticks and start with the letter r before the actual code (to indicate that this is R code).

solution

The inline code you can use for this is the following (remember to add the letter r between the opening backtick and the command in your document): nrow(corona_survey) corona_survey %>% filter(sex == 2) %>% nrow() corona_survey %>% filter(sex == 1) %>% nrow() Repeat this for the reduced data set (which should be called corona_survey_noncrit).

6

In the Measures (sub)section you can describe which variables we will use for our analysis and how they were measured. The predictor variables are risk_self, risk_surroundings, and risk_infect_others, and the outcome/target variable is obey_curfew.

For describing the measures, you can consult the codebook for the original data set. The original variable names are: hzcy001a, hzcy002a, hzcy005a (predictors) and hzcy026a (outcome).

7

For the part on Descriptive statistics in the Results section, generate a summary table for the risk_ variables using the stargazer package. For all analyses, we want to use the data from the reduced/filtered data set (which should be called corona_survey_noncrit).

For the selection of variables you can use one of the helper functions for dplyr::select(). Remember that the stargazer() function does not work with tibbles and expects a dataframe. Since we want to create HTML output, you should specify “html” as the type for the stargazer table. For prettier results, you may also want to give the table a title, specify that you want two decimal places, and also change the variable names (not in the data set but in the pipe for generating the table). To have the table properly displayed in the output document, you need to set the chunk option results = 'asis'.

solution 

corona_survey_noncrit %>% 
  select(starts_with("risk")) %>% 
  rename(`Own infection` = risk_self, # normally spaces in variable names are bad practice, but here it is only for the tabular output
         `Infection in close surroundings` = risk_surroundings,
         `Infecting others` = risk_infect_others) %>% 
  as.data.frame() %>% 
  stargazer(type = "html",
            digits = 2,
            title="Table 1. Descriptive statistics")

8

As the next analysis step, we want to compute the correlations between the risk variables and produce a table displaying them. For that, you can use the corrr package (for computing the correlations) and combine it with a function from the knitr package for creating the table.

We can use the same pipe steps as before for selecting the variables and optionally renaming them. The main workhorse of the corrr package is its correlate() function. The knitr function for creating tabular output is kable(). To make the output nicer to read, we can use two additional functions from the corr package for removing double entries (i.e., the upper triangle) from the correlation matrix and cleanly formatting it. You can consult corr package documentation to find out how to do this. In addition, you might want to add a caption and change the columns names via two arguments in the knitr::kable() function.

solution 

corona_survey_noncrit %>% 
  select(starts_with("risk")) %>% 
  rename(`Own infection` = risk_self,
         `Infection in close surroundings` = risk_surroundings,
         `Infecting others` = risk_infect_others) %>% 
  correlate() %>% 
  shave() %>% 
  fashion(na_print = "—") %>% 
  knitr::kable( # pkg_name::function is a way of using functions from a package without loading it
    caption = "Table 2. Perceived personal likelihoods of infection with & transmission of the Corona virus",
    col.names = c("Measure", "1", "2", "3")
  )

9

Our main analysis for this report is a logistic regression model with the risk variables as predictors and the obey_curfew variable as the outcome. Let’s first compute this simple model before we produce a table to present its results.

If you have not done a logistic regression in R yet, a simple web search should quickly turn up many useful results that can show you how do to this. As a hint or reminder, you need the glm() function which expects the first argument (i.e., the regression formula) in the same format as the lm() function. However, in addition to that, you need to specify that you want to use a logit link (via the family argument).

solution 

model <- glm(obey_curfew ~ risk_self + risk_surroundings + risk_infect_others, 
             family = binomial(link = "logit"),
             data = corona_survey_noncrit)

10

Now that we have computed the logistic regression model, let’s create a table to display its results. You can use the stargazer package for this in a similar way as for the summary stats table.

As its first argument, for regression tables, the stargazer() function expects one or more model objects. For a prettier table, you can also specify labels for the dependent variables and the predictors. Have a look at the help file for the stargazer() function via ?stargazer() to see how to do this. Also here, for correct printing in the output document you need to set the chunk option results = 'asis'.

solution 

stargazer(model,
          type = "html",
          dep.var.labels=c("Willingness to comply with curfew"),
          covariate.labels=c("Risk of own infection",
                             "Risk of infection in close surroundings",
                             "Risk of infecting others"),
          title="Table 3. Results of the logistic regression model")

11

Write one or two sentences describing the key results of the logistic regression model in the Discussion section.

You can run the code for the logistic regression model (without assigning the result to an object) in your console to check out its results. Of course, before doing that, you also need to run the wrangling code.

12

To further increase reproducibility, let’s add some information about your R session (OS, R version, loaded packages) to the document. Add another section at the end of the document of the document called Reproducibility information and display this information there.

You can use a base R function to access info about your session (function name hint 1) and another one for printing it (function name hint 2). You can exclude information about your locale settings there.

solution 

sessionInfo() %>% 
  print(locale = FALSE)

13

After having created the R Markdown document, let’s knit the HTML output. You should save the resulting .html file in the output folder within your project repository.

You can either simply use the Knit button in the RStudio GUI and then (manually) copy the resulting file into the output folder (by default, output documents are stored in the same location as the source file). Alternatively, you can use the render() command from the rmarkdown package and provide it with the name/location of the file you want to knit and a path for the output file. For the second option, make sure that your working directory is set appropriately (or adjust the file paths in the solution accordingly).

solution 

rmarkdown::render('Report1_Obeying_Curfew.Rmd', output_file = '../output/Report1_Obeying_Curfew.html')

Finally, as before, after you have created and saved the .Rmd document and the HTML output, in Git, add the files, commit them (with a meaningful commit message), and push the changes to your GitHub repository.